BEGIN:VCALENDAR
VERSION:2.0
PRODID:talks.ox.ac.uk
BEGIN:VEVENT
SUMMARY:Generics and Default Reasoning in Large Language Models  - Ethics 
 in AI Lunchtime Research Seminars - Dr James Kirkpatrick (Philosophy\, Uni
 versity of Oxford)
DTSTART;VALUE=DATE-TIME:20250226T123000Z
DTEND;VALUE=DATE-TIME:20250226T133000Z
UID:https://talks.ox.ac.uk/talks/id/353eac80-eb89-4414-b553-30a0104e0785/
DESCRIPTION: In recent years\, there has been growing interest in the reas
 oning abilities of Large Language Models (LLMs). However\, there is relati
 vely little discussion on how LLMs handle default reasoning patterns that 
 have motivated various systems of non-monotonic logic. This talk examines 
 the capabilities of 28 LLMs to reason with 13 reasoning patterns widely di
 scussed in the non-monotonic logic literature\, focusing on inference patt
 erns involving generics (e.g.\, `Birds fly'\, `Ravens are black') that exp
 ress default information. Generics are notable because they express quasi-
 universal generalisations that exhibit complex exception-permitting behavi
 our that cannot easily be described in set-theoretic or Bayesian terms. Ge
 nerics are of special interest to philosophers\, linguistics\, logicians\,
  and cognitive scientists\, not least due to their connection with default
  reasoning but because of their centrality in cognition and concept acquis
 ition. Assessing the capacity of LLMs for default reasoning is important f
 or evaluating the extent to which the reasoning abilities of LLMs match th
 ose of humans more generally.\nSpeakers:\nDr James Kirkpatrick (Philosophy
 \, University of Oxford)
LOCATION:Please register to receive venue details
TZID:Europe/London
URL:https://talks.ox.ac.uk/talks/id/353eac80-eb89-4414-b553-30a0104e0785/
BEGIN:VALARM
ACTION:display
DESCRIPTION:Talk:Generics and Default Reasoning in Large Language Models  
 - Ethics in AI Lunchtime Research Seminars - Dr James Kirkpatrick (Philoso
 phy\, University of Oxford)
TRIGGER:-PT1H
END:VALARM
END:VEVENT
END:VCALENDAR